Upload
others
View
11
Download
0
Embed Size (px)
Citation preview
A flexibly linkable meta layer of geographic features supplementary for
driving automation and simulation
DSC 2020 EuropeVR
Rüdiger Ebendt
> A flexibly linkable meta layer of geographic features > R. Ebendt > 10.09.2020DLR.de • Chart 1
Motivation 1(2)
• High Definition (HD) Maps: which is the “best” format?
• OpenDRIVE
• IPG ROAD5
• NDS Open Lane Model (Navigation Data Standard)
• Lanelet/Lanelet2
• TomTom HD Map RoadDNA
• Here HD Live Map
• …or even OpenStreetMap (OSM)?
• Open source maps and community crowdsourcing approaches are emerging with OpenStreetMaps
(OSM)
• Some vendors and a rising number of start-ups use them as a basis for navigable maps with
additional navigation-related attributes generated via AI tools
• Scientific progress in the use of OSM data for global path planning or (even small scale) localization
• This may also put Standard Definition (SD) maps from Here, TomTom etc. back into play
> A flexibly linkable meta layer of geographic features > R. Ebendt > 10.09.2020DLR.de • Chart 2
Motivation 2(2)
• Lack of standards: Despite ongoing standardization efforts such as NDS, ADASIS, SENSORIS, TISA, etc.,
which are overlooked by OADF, “maps are still essentially proprietary datasets lacking interoperability
between mapping suppliers” (ABI Research / HERE, 2018).
• Even if a “best” standard could be identified: With respect to what purpose, such as driving automation /
ADAS or driving simulation, would it be “best”?
• Example: Different requirements on relative and absolute errors for simulation and ADAS
• Different types of map content:
• 3D representations of
• buildings and various types of landmarks
• slope and curvature of roads, lane markings, and roadside objects, such as sign posts
• Street level, aerial, and satellite imagery allowing to derive
• green spaces and lane boundaries
➢Many different data sources for continuous updates and conflation into different layers of the HD map
> A flexibly linkable meta layer of geographic features > R. Ebendt > 10.09.2020DLR.de • Chart 3
Idea of the data model / meta layer „Road2Automation“ (R2A) 1(2)
The aim is to
• supplement road information in today’s digital road maps with georeferenced features which are relevant
to
• driving automation and
• driving simulation
• address maps of diverse levels of detail, precision and format (ranging from HD and SD maps from different
vendors to crowd-sourced data from OpenStreetMap (OSM))
• facilitate transfer of the features between maps, based on different data sources by a meta layer on top
of all source and target maps, together with appropriate runtime modules for location referencing. This
supports
• map conflation and continuous map updates
• use cases, where information from more than one type of map / more than one data source is
required.
> A flexibly linkable meta layer of geographic features > R. Ebendt > 10.09.2020DLR.de • Chart 4
Idea of the data model / meta layer „Road2Automation“ (R2A) 2(2)
> A flexibly linkable meta layer of geographic features > R. Ebendt > 10.09.2020DLR.de • Chart 5
Different map formats (HD and SD formats)
R2A:
Data model
and meta-layer
…
R2A is a „hybrid“ of
• a data model
• a meta layer on top
of „source maps“
References
via
• flexible reference ID
• OpenLR-reference
• Road network matching
Examples for use cases addressed by R2A 1(4)
A: Driving automation
• Global path planning & (large and even small scale) localization
• Relevant features (already available in SD maps):
• name, type and width of the streets,
• public speed limits
• (OSM) track data
• topology of the road network
• Safety of autonomous transport
• Relevant features:
• Information about complex intersections with attributes indicating e.g.
• poor visibility of arms
• absence of traffic lights
• presence of many fast or wide lanes,
• presence of short green light periods,
• accident hotspots
• Dangerous turns or bends
> A flexibly linkable meta layer of geographic features > R. Ebendt > 10.09.2020DLR.de • Chart 6
Examples for use cases addressed by R2A 2(4)
• Safety of autonomous transport (continued)
• Relevant features:
• horizontal and vertical curves with their radius and pitch
• Information with relevance to the sensor system:
• quality of lane markings
• presence of tunnels or guardrail anti-glare panels
• presence of steaming manholes
• “Transition areas” = lane sections, where either smart infrastructure has to exist or the driver must take over
driving upon entering
• Examples:
• work zones
• merging areas, lane drops
• no automation zones
> A flexibly linkable meta layer of geographic features > R. Ebendt > 10.09.2020DLR.de • Chart 7
Examples for use cases addressed by R2A 3(4)
• Energy efficient driving (by acceleration optimization)
• Relevant features:
• intersections
• slopes
• presence of tunnels
• Curve or hill warnings
• Relevant features:
• curves (radius / pitch)
• slopes
• Predictive cornering light
• Relevant features:
• curve radius
> A flexibly linkable meta layer of geographic features > R. Ebendt > 10.09.2020DLR.de • Chart 8
Source: Volkswagen Communications,
Arteon assistance systems, “Active
Lighting System”
Examples for use cases addressed by R2A 4(4)
B: Driving simulation, especially for testing partial vehicle automations / ADAS
• Same situation as for driving automation: use of many different data sources
• R2A provides a meta layer for the linkage between possible different sources for the static content of a
scenario
> A flexibly linkable meta layer of geographic features > R. Ebendt > 10.09.2020DLR.de • Chart 9
Source: based on Daniel Lipinski, „Wie gut ist gut genug?“,
Closing event of project PEGASUS
R2A meta layer
In order to establish a reference to the location of an
R2A feature in a source map, R2A links back to:
R2A data model
Feature attributes in the model are populated by:
The two purposes of R2A
• Features in HD maps, often obtained by terrestial
mobile mapping (in OpenDRIVE or
Road2Simulation format)
• Road reference lines (Road2Simulation)
• Roads and lanes (OpenDRIVE)
• Junctions or junction groups (OpenDRIVE)
• Features in SD maps from OpenStreetMap, HERE,
or TomTom
• Road segments (e.g. HERE (GDF) edges, or
OSM ways)
• Attributes originally contained in the (HD or SD)
maps
• Road topology, road geometry, tunnels, lane
markings, …
• Attributes filled with data from data collections (e.g.
by mobile mapping, remote sensing techniques
such as image flight) or with data derived by post-
processing
• Accident hotspots, poor visibility of junction
arms, curve radius or pitch
• Quality of lane markings
• Locations of road side units, blue tooth sensors,
landmarks, …
> A flexibly linkable meta layer of geographic features > R. Ebendt > 10.09.2020DLR.de • Chart 10
Class diagram of R2A model 1(2)
> A flexibly linkable meta layer of geographic features > R. Ebendt > 10.09.2020DLR.de • Chart 11
GuardrailAntiglare,
MarkQuality,
AverageMarkQuality,
Curve,
…
LandMark,
SteamingManhole,
RoadSideUnit,
…
Class diagram of R2A model 2(2)
> A flexibly linkable meta layer of geographic features > R. Ebendt > 10.09.2020DLR.de • Chart 12
3D geometry, for example
• Point Z (605044.419819 5791949.77898 128.12)
or
• LineString Z (605059.7409 5791956.6 128.12,
605014.3896 5791935.49 128.22)
Base64 encoded OpenLR string, for example
CwOiYCUMoBNWAv9P/+MSBg==
Reference identifier of a default type (StringId)
Intended default use: reference to an OpenDRIVE
<junctionGroup>
Flexible reference identifiers 1(2)
> A flexibly linkable meta layer of geographic features > R. Ebendt > 10.09.2020DLR.de • Chart 13
…
…
Flexible reference identifiers 2(2)
• Design principles:
• Abstract classes which force derived concrete classes to implement a common interface (for instance
the getter and setter for an attribute refId)
• This allows for generic type-agnostic code as well as for dynamic method dispatch based on the
actual key type as given by the static attribute ID_TYPE
• This supports the use of runtime polymorphism and of object oriented (Java) design patterns
> A flexibly linkable meta layer of geographic features > R. Ebendt > 10.09.2020DLR.de • Chart 14
More detailed:
Transfer of features between dissimilar maps: R2A run time modules for
inter-map matching (work in progress)
maps
> A flexibly linkable meta layer of geographic features > R. Ebendt > 10.09.2020DLR.de • Chart 15
OSM (purple) and HERE (orange) maps of Berlin, Germany Made with QGIS (https://qgis.org)
Road network matching and two applications (terminology of M. Zhang, PhD
thesis 2009)
> A flexibly linkable meta layer of geographic features > R. Ebendt > 10.09.2020DLR.de • Chart 16
a) Road network matching b) Road network integration c) Road network conflation
(often also called map
conflation)Source: Zhang, M.: Methods and Implementations of Road-Network Matching, Ph.D. thesis
Dynamic location referencing by OpenLR: line locations (routes)
> A flexibly linkable meta layer of geographic features > R. Ebendt > 10.09.2020DLR.de • Chart 17
• OpenLR is a bandwidth-efficient protocol for the transfer of
• linear locations
• area locations
from a sender to a receiver
• They are encoded with the map of the sender and decoded with the
map of the receiver
• Idea:
• Instead of transmitting every waypoint of the source route,
cover it by a concatenation of shortest paths between intermediate
location reference points (LRP)
• Transmit only these LRP
Dynamic location referencing by OpenLR: OpenLR: area locations
> A flexibly linkable meta layer of geographic features > R. Ebendt > 10.09.2020DLR.de • Chart 18
Road network matching with the Geometry Inter-Map Matching Extension
(GIMME)
1. Geometry Matching (GM) orders candidate edges with respect to their “fitness” as a target edge matching a
source edge
2. GIMME (R. Ebendt, L.C. Touko Tcheumadjeu, Eur. Transp. Res. Rev. 9, 38, 2017) optimizes the target
route, starting on a collection of ordered candidate lists for every source edge
> A flexibly linkable meta layer of geographic features > R. Ebendt > 10.09.2020DLR.de • Chart 19
Source: based on Robert Sämann, master thesis, 2014
OpenLR Road network matchers
Comparison of OpenLR and road network matchers like GIMME
• Success rates of more than 90% when matching
maps from different vendors, and 98-99%, when
matching different map releases of the same vendor
• Bandwidth-efficient (by transmitting only as much
LRP as needed)
• Reference implementation of OpenLR is available
as open source
• Map-agnostic: sender does not need to know
which map is used at the receiver side
• Very high success rates (GIMME: 99.7%) when
matching maps from different vendors, and almost
perfect results (very close to 100%) when matching
different map releases of the same vendor
• Bandwidth-consuming1) since complete
geometries (line strings) must be transferred from
sender to receiver
• No actively maintained open source
implementations seem to be available
• Map-independent (works with any map pair from
different vendors like HERE, TomTom,
OpenStreetMap, …), but receiver needs to have
access to the same map as the sender
> A flexibly linkable meta layer of geographic features > R. Ebendt > 10.09.2020DLR.de • Chart 20
1) bandwidth-efficient, when combined with a bandwidth-
efficient method, but at the cost of additional runtime
overhead
Conclusion
• A data model and meta layer called „Road2Automation“ (R2A) has been presented
• It addresses several use cases in driving automation and driving simulation
• Its main characteristics are
• flexibly extensible design: the meta layer can establish links between maps or data sets in virtually any
format
• support of feature transfer between maps (as required for map conflation, map integration, map
updates, as well as for some of the addressed use cases)
• Run time modules for inter-map matching are required alongside the model
• Adaption of already existing such modules (OpenLR, GIMME) to R2A is work in progress
• This work is currently done in two institutionally funded projects of German Aerospace Center
• KoFiF („Kooperative Fahrzeugintelligenz und mechatronisches Fahrwerk“), subproject of the large scale
project Next Generation Car (NGC)
• Cross-sectoral project “Digital Atlas”
> A flexibly linkable meta layer of geographic features > R. Ebendt > 10.09.2020DLR.de • Chart 21
> A flexibly linkable meta layer of geographic features > R. Ebendt > 10.09.2020DLR.de • Chart 22
Source: Nate Grigg/Flickr